duration of Disaggregation

[Bart Vleminckx]

Dear all,

We ran the disaggregation for a large hazard calculation, after a few days we reached 1%... (with 16 cores). So we did a small test in which the calculation of the hazard curve takes about 30 seconds and the the disaggregation part 6 minutes. When we set the bin parameters that there is only one bin for magnitude, distance, coordinates and epsilon (so no disaggregation), the disaggregation part still takes 5 minutes... How is this possible?

Best regards,

Bart Vleminckx

Royal Observatory of Belgium

[Damiano Monelli]

Hi Bart,

the reason why hazard curves calculation is much faster than disaggregation is because the first is parallelized over sources, while the

second is parallelized over sites. In other words, for calculating a disaggregation matrix for a site, calculations for all sources are done in

a single task.

Having said that, we have also experienced very long computation times associated with a disaggregation calculation, and we are planning

to revise it to improve the situation.

In the meantime I would suggest you to simplify as much as possible the job.ini file to reduce the number of calculations requested (that is split the 

big calculation into a set of smaller calculations). For instance just consider a single IMT and a single POE. If you are interested in only magnitude-distance-

epsilon disaggregation just put very large values for the 'coordinate_bin_width' so that you reduce the size of the full disaggregation matrix.

Best regards,

Damiano

--
You received this message because you are subscribed to the Google Groups "OpenQuake Developers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openquake-de...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.